12 March 2025
IP25030 | What Will Trump 2.0 Do? The AI Diffusion Rule and DeepSeek’s Implications for Southeast Asia
SYNOPSIS
The Biden administration’s “parting gift” of the AI Diffusion Rule expands US chip controls on China from “small yards, high fences” to an interventionist global strategy with significant repercussions for Southeast Asia’s AI aspirations and economic development. Trump 2.0’s response to the AI Diffusion Rule will have to account for DeepSeek’s disruption of the AI landscape. While innovations such as DeepSeek and open-source AI architectures offer long-term potential for regional development, in the short term, Southeast Asian countries will have to be cautious as they navigate the US-China technology competition.
COMMENTARY
In its waning days, the Biden administration announced a new “Framework for Artificial Intelligence Diffusion” (the AI Diffusion Rule) that transformed the United States’ approach to denying China access to cutting-edge AI technologies from “small yard, high fences” into an expansive and interventionist strategy that privileges Washington as the “gatekeeper” for global AI development.
The AI Diffusion Rule creates a three-tier classification system for the export of advanced AI chips. Countries in Tier-1 comprise 18 close US allies, including the members of the “Five Eyes” intelligence alliance, vital partners in the AI supply chain (Japan, South Korea, the Netherlands, and Taiwan), and some NATO allies (such as Germany, France, and Italy). These countries face essentially unrestricted access to United States’ AI technologies. In Tier-3, adversaries such as China, Russia, North Korea, Iran, and “other countries of concern” remain subject to near-total embargoes on advanced AI hardware and software.
This leaves most countries — including Singapore and other Southeast Asian countries — in Tier-2 and facing hard limits on the total processing power (TPP) that they can import, which translates to around 50,000 Nvidia H100s through 2027. Approved US-headquartered Tier-1 technology companies (i.e., major US hyperscalers) are also limited to deploying not more than 7 per cent of their AI compute capacity in any single Tier-2 country out of an allocation of 25 per cent total compute capacity.
Companies in Tier-2 countries can apply for a National Validated End User (NVEU) authorisation if they seek to import a significant number of GPUs. The regulations “strongly encourage” these companies to first secure a government-to-government assurance before seeking US regulatory approval, and successful applicants will need to comply with policy and regulatory requirements, as well as sever supply chain and other technology ties with China. These larger allocations of chips, however, will remain subject to stringent per-company and per-country quarterly limits on compute power. The strict quotas on Tier-2 countries are designed to prevent them from importing the compute necessary for training and running the most advanced AI models, ensuring they lag behind the Tier-1 countries at frontier.
What Will Trump 2.0 Do?
The AI Diffusion Rule is scheduled to go into effect on 15 May 2025 after the 120-day comment period ends. Given the expected loss of revenue and potential market share to Chinese alternatives, it is unsurprising that the US technology sector’s reaction to the regulations has been extremely hostile, with Nvidia accusing it of “sweeping overreach” that will undermine American leadership in AI and Oracle lambasting it as “the most destructive to ever hit the US technology industry”.
It remains to be seen, however, whether the Trump administration will seek to implement the AI Diffusion Rule. The first Trump administration initiated technology restrictions against Huawei and ZTE and there is no shortage of China hawks in the second administration. There are reports that the administration is contemplating further tightening of chip controls. However, US president Donald Trump also has a history of reversing the decisions of his Democrat predecessors. He has issued a directive calling for a review of the export control system and recommendations on maintaining US technology leadership and eliminating loopholes, and also revokedJoe Biden’s executive order reducing the risks that AI poses to national security, the economy, and public health or safety. Where President Trump’s pick for AI and crypto “Czar”, David Sacks, stands on the AI Diffusion Rule is undetermined, and in his confirmation hearings, Commerce Secretary Howard Lutnick supportedcontinued chip controls but also acknowledged their limits given DeepSeek’s AI breakthrough.
DeepSeek and AI’s “Linux Moment”? Implications for Southeast Asia
On the day Trump took office for his second term, Hangzhou-based AI startup DeepSeek released an open-source AI model whose performance rivalled those of US tech giants such as OpenAI’s ChatGPT but which was cost and compute efficient. DeepSeek’s disruption proved both the effectiveness and the limits of US export controls: faced with the scarcity of embargoed advanced Nvidia AI chips, the company adapted and innovated to seek algorithmic efficiency and specialised architectures to produce competitive AI models reportedly costing only a fraction of those of US rivals. DeepSeek’s success has predictably resulted in calls for banning it but Trump’s response was measured and pragmatic, calling it a “wake-up call” and recognising that American AI companies could also exploit its breakthrough to become more efficient.
The AI Diffusion Rule will leave Southeast Asian countries disadvantaged in their ability to develop or deploy frontier AI models, with repercussions for their economic and social development and regional aspirations to become a digital hub. The DeepSeek “Linux Moment” — offering an affordable and efficient open-source alternative to expensive and environmentally impactful compute-intensive strategies — holds the promise of democratising access to AI, allowing less-developed economies, as well as small, medium, and micro enterprises (SMMEs) and startups, access to cutting-edge technologies without breaking the bank. It could also enhance flexibility as open-source models allow regional governments, businesses, and developers to modify AI models to fit local needs and conditions — from diverse linguistic requirements to region- or country-specific challenges. It could also spur regional AI collaboration and digital connectivity — in line with ASEAN’s ambitions of digital integration under initiatives such as its Digital Economy Framework Agreement (DEFA).
While a potentially transformative opportunity, DeepSeek and open-source AI are not a silver bullet and come with downsides. Open-source AI still requires significant technical knowledge that may be scarce in countries with less developed tech ecosystems. Further, while open-source AI can reduce costs, investments in digital infrastructure are still essential to ensure stable and widespread connectivity.
There are security risks associated with both DeepSeek and open-source AI. With DeepSeek, there are security and privacy risks — quite aside from its tendency to evade answering “sensitive” questions on topics often censored by the Chinese government. South Korea, Italy, and Australia are among countries that have blocked access to DeepSeek, citing security concerns such as the lack of safeguards to mitigate risk of exploitation as well as storage of user data in China.
DeepSeek is also likely to add fuel to the fire of the US-China technology war. Beijing has touted DeepSeek as a “win”, buoying its belief that it can out-innovate the United States despite Washington’s export controls. DeepSeek has already put pressure on its larger domestic competitors by becoming the new benchmark. Alibaba released its Qwen 2.5-Max model, which it claims outperforms DeepSeek-V3, while Tencent released its Hunyuan Turbo S that it claims can answer queries faster than DeepSeek’s R1 and matches V3 when tested on fields like knowledge, math, and reasoning. This will inevitably provoke American countermeasures.
Conclusion
In the short term, Southeast Asian countries will have to be cautious while adopting DeepSeek or open-source AI systems that may appear to Washington as attempts to break away from the US tech ecosystem — especially as they face the threat of reciprocal tariffs and an impending trade war. Over the longer term, however, regional investments in open-source AI architectures — seizing upon the “Linux Moment” — can help maintain strategic balance by enhancing autonomy and reducing dependencies on either Washington or Beijing.
Adrian Ang U-Jin is Research Fellow and Coordinator of the United States Programme at the S. Rajaratnam School of International Studies (RSIS).
SYNOPSIS
The Biden administration’s “parting gift” of the AI Diffusion Rule expands US chip controls on China from “small yards, high fences” to an interventionist global strategy with significant repercussions for Southeast Asia’s AI aspirations and economic development. Trump 2.0’s response to the AI Diffusion Rule will have to account for DeepSeek’s disruption of the AI landscape. While innovations such as DeepSeek and open-source AI architectures offer long-term potential for regional development, in the short term, Southeast Asian countries will have to be cautious as they navigate the US-China technology competition.
COMMENTARY
In its waning days, the Biden administration announced a new “Framework for Artificial Intelligence Diffusion” (the AI Diffusion Rule) that transformed the United States’ approach to denying China access to cutting-edge AI technologies from “small yard, high fences” into an expansive and interventionist strategy that privileges Washington as the “gatekeeper” for global AI development.
The AI Diffusion Rule creates a three-tier classification system for the export of advanced AI chips. Countries in Tier-1 comprise 18 close US allies, including the members of the “Five Eyes” intelligence alliance, vital partners in the AI supply chain (Japan, South Korea, the Netherlands, and Taiwan), and some NATO allies (such as Germany, France, and Italy). These countries face essentially unrestricted access to United States’ AI technologies. In Tier-3, adversaries such as China, Russia, North Korea, Iran, and “other countries of concern” remain subject to near-total embargoes on advanced AI hardware and software.
This leaves most countries — including Singapore and other Southeast Asian countries — in Tier-2 and facing hard limits on the total processing power (TPP) that they can import, which translates to around 50,000 Nvidia H100s through 2027. Approved US-headquartered Tier-1 technology companies (i.e., major US hyperscalers) are also limited to deploying not more than 7 per cent of their AI compute capacity in any single Tier-2 country out of an allocation of 25 per cent total compute capacity.
Companies in Tier-2 countries can apply for a National Validated End User (NVEU) authorisation if they seek to import a significant number of GPUs. The regulations “strongly encourage” these companies to first secure a government-to-government assurance before seeking US regulatory approval, and successful applicants will need to comply with policy and regulatory requirements, as well as sever supply chain and other technology ties with China. These larger allocations of chips, however, will remain subject to stringent per-company and per-country quarterly limits on compute power. The strict quotas on Tier-2 countries are designed to prevent them from importing the compute necessary for training and running the most advanced AI models, ensuring they lag behind the Tier-1 countries at frontier.
What Will Trump 2.0 Do?
The AI Diffusion Rule is scheduled to go into effect on 15 May 2025 after the 120-day comment period ends. Given the expected loss of revenue and potential market share to Chinese alternatives, it is unsurprising that the US technology sector’s reaction to the regulations has been extremely hostile, with Nvidia accusing it of “sweeping overreach” that will undermine American leadership in AI and Oracle lambasting it as “the most destructive to ever hit the US technology industry”.
It remains to be seen, however, whether the Trump administration will seek to implement the AI Diffusion Rule. The first Trump administration initiated technology restrictions against Huawei and ZTE and there is no shortage of China hawks in the second administration. There are reports that the administration is contemplating further tightening of chip controls. However, US president Donald Trump also has a history of reversing the decisions of his Democrat predecessors. He has issued a directive calling for a review of the export control system and recommendations on maintaining US technology leadership and eliminating loopholes, and also revokedJoe Biden’s executive order reducing the risks that AI poses to national security, the economy, and public health or safety. Where President Trump’s pick for AI and crypto “Czar”, David Sacks, stands on the AI Diffusion Rule is undetermined, and in his confirmation hearings, Commerce Secretary Howard Lutnick supportedcontinued chip controls but also acknowledged their limits given DeepSeek’s AI breakthrough.
DeepSeek and AI’s “Linux Moment”? Implications for Southeast Asia
On the day Trump took office for his second term, Hangzhou-based AI startup DeepSeek released an open-source AI model whose performance rivalled those of US tech giants such as OpenAI’s ChatGPT but which was cost and compute efficient. DeepSeek’s disruption proved both the effectiveness and the limits of US export controls: faced with the scarcity of embargoed advanced Nvidia AI chips, the company adapted and innovated to seek algorithmic efficiency and specialised architectures to produce competitive AI models reportedly costing only a fraction of those of US rivals. DeepSeek’s success has predictably resulted in calls for banning it but Trump’s response was measured and pragmatic, calling it a “wake-up call” and recognising that American AI companies could also exploit its breakthrough to become more efficient.
The AI Diffusion Rule will leave Southeast Asian countries disadvantaged in their ability to develop or deploy frontier AI models, with repercussions for their economic and social development and regional aspirations to become a digital hub. The DeepSeek “Linux Moment” — offering an affordable and efficient open-source alternative to expensive and environmentally impactful compute-intensive strategies — holds the promise of democratising access to AI, allowing less-developed economies, as well as small, medium, and micro enterprises (SMMEs) and startups, access to cutting-edge technologies without breaking the bank. It could also enhance flexibility as open-source models allow regional governments, businesses, and developers to modify AI models to fit local needs and conditions — from diverse linguistic requirements to region- or country-specific challenges. It could also spur regional AI collaboration and digital connectivity — in line with ASEAN’s ambitions of digital integration under initiatives such as its Digital Economy Framework Agreement (DEFA).
While a potentially transformative opportunity, DeepSeek and open-source AI are not a silver bullet and come with downsides. Open-source AI still requires significant technical knowledge that may be scarce in countries with less developed tech ecosystems. Further, while open-source AI can reduce costs, investments in digital infrastructure are still essential to ensure stable and widespread connectivity.
There are security risks associated with both DeepSeek and open-source AI. With DeepSeek, there are security and privacy risks — quite aside from its tendency to evade answering “sensitive” questions on topics often censored by the Chinese government. South Korea, Italy, and Australia are among countries that have blocked access to DeepSeek, citing security concerns such as the lack of safeguards to mitigate risk of exploitation as well as storage of user data in China.
DeepSeek is also likely to add fuel to the fire of the US-China technology war. Beijing has touted DeepSeek as a “win”, buoying its belief that it can out-innovate the United States despite Washington’s export controls. DeepSeek has already put pressure on its larger domestic competitors by becoming the new benchmark. Alibaba released its Qwen 2.5-Max model, which it claims outperforms DeepSeek-V3, while Tencent released its Hunyuan Turbo S that it claims can answer queries faster than DeepSeek’s R1 and matches V3 when tested on fields like knowledge, math, and reasoning. This will inevitably provoke American countermeasures.
Conclusion
In the short term, Southeast Asian countries will have to be cautious while adopting DeepSeek or open-source AI systems that may appear to Washington as attempts to break away from the US tech ecosystem — especially as they face the threat of reciprocal tariffs and an impending trade war. Over the longer term, however, regional investments in open-source AI architectures — seizing upon the “Linux Moment” — can help maintain strategic balance by enhancing autonomy and reducing dependencies on either Washington or Beijing.
Adrian Ang U-Jin is Research Fellow and Coordinator of the United States Programme at the S. Rajaratnam School of International Studies (RSIS).
